Issue



How hot is the thermal equipment Market


02/01/1997







How hot is the thermal equipment market?

John M. Salzer, Salzer Technology Enterprises Inc., Santa Monica, California

This discussion of the thermal market will focus on basic thermal applications - equipment applied for heating without deposition. Furnaces used for diffusion, oxidation, annealing, and the like are included, but not when used for deposition. Similarly, rapid thermal processors (RTP) are included when employed in diffusion, oxidation, nitridation, annealing, silicidation, glass reflow, etc., but not in CVD or epitaxy.

RTP is the new boy on the block and has successfully carved out a market niche, but furnace makers have not stood by. Furnaces have a long history, which includes constant improvements and changes and an industry-wide infrastructure, background, training, and know-how.

Furnaces were already used for diffusion to form junctions of the first discrete transistors. Much later, ion implantation was introduced, and it looked for a while that in junction formation it might displace furnaces. Instead, although ion implanters have come to play a major role in semiconductor fabrication, furnaces continue to be used for the same purpose. Furthermore, furnaces or RTP systems are used for annealing to remove damage and to activate dopants after ion implantation.

I foresee a similar situation in the thermal market, in which use of RTP will increase without replacing furnaces altogether. The furnace industry is agile - its switch from horizontal to vertical furnaces provided a very pivotal improvement, and in this case the vertical configuration significantly displaced the horizontal one.

Figure 1. Comparative ranges of ramp rates of thermal systems.

Conventional furnaces use a large batch of wafers, numbering between 100 and 200, and can provide good throughput, particularly when the process requires a long time at temperature. In such a case, the time delay in heating up to the process temperature and cooling down to "ambient" is a smaller fraction of the total time consumed than if the process is a short one, and the long ramp time becomes a major part of the time required.

In the latter case, RTP may be the logical solution. However, there are many other reasons for RTP usage. RTP consumes less thermal budget for the same job, is a single-wafer (in one case, two-wafer) system with competitive throughput, and can turn a heat source, and thus the process, on and off instantly to achieve rapid sequencing to critical thicknesses. RTP is also compatible with cluster configurations, either for a multiplicity of RTP chambers to increase throughput or as part of an integrated process sequence. When the thermal process is part of an integrated sequence, RTP obviates exposing the wafer to particulates during an intertool transfer.

Furnaces have difficulty in performing some applications. For example, titanium silicidation is very sensitive to oxygen and vapor, which are difficult to eliminate in furnaces. This step, which is not very sensitive to temperature control, has become a major application of RTP.

The broader use of RTP has been delayed by the difficulty of controlling temperature. Furnaces are basically isothermal: the walls and wafers are in total temperature equilibrium. In contrast, various parts of an RTP chamber are at different temperatures, so it is more difficult to achieve uniformity over a wafer or repeatability from wafer to wafer. Furthermore, the emissivity of the back surface of the wafer measured by a pyrometer (the most common temperature control method) varies with the character of the surface, due to the degree of roughness or the deposited materials.

Recent solutions to the problem of temperature measurement have expanded the applicability of RTP. As a result, the RTP market grew very rapidly in 1995, surpassing even the stellar performance of the overall equipment market. However, whereas the continued growth of the RTP industry has never been in question, the 1996 weakness in the semiconductor industry had a serious effect. Not only did the device makers buy less equipment, but they were also less likely to switch to a different thermal system.

And now the furnace makers raised another challenge - the minibatch, fast-ramp furnace. These furnaces may hold only 25-50 wafers, spaced further apart than those in the conventional furnaces, but the heating system has been strengthened so that ramping takes much less time.

Ramping and throughput

Ramp-up can easily take an hour in a conventional furnace, compared to 10-20 minutes in a fast-ramp model, and just 10-60 seconds in an RTP system. Ramp rates in practical systems are probably a more meaningful indicator and are 10-20?C/min, 30-50?C/min, and 300-1200?C/min, respectively. RTP rates are usually specified in terms of ?C/sec and may range from 50 to 200?C/sec or even more. These ramp rates are illustrated in Fig. 1, whose y axis is logarithmic.

To use throughput alone as a basis of comparison, we have to define the particular applications. In general, conventional furnaces have the highest throughput - about 60-100 wafers/hour - compared to the other two types of thermal systems. Simply adding the up and down ramp times plus the process times is not sufficient. One must also consider the time spent in preparing a wafer load and loading it. Full-size furnaces counteracted these delays by clustering two vertical tubes so that one is processing while the other one is being loaded. Of course, the price goes up with this duplication and with whatever elaborate loading automation is included.

It is a bit early to determine the practical throughput of fast-ramp furnaces as they are just starting to be put to use. While raw throughput may be in the range of 40-70 wafers/hour, we have to await their fab-line performances. The claims of their manufacturers are confident and, indeed, promising. At this stage, they appear to have a growing role to play and may take market share away from either the large furnaces or RTP systems or both.

RTP systems have come into their own in the last few years as their heating and temperature measuring systems reached a higher level of sophistication. In addition to silicidation, mentioned before, RTP has been used in annealing after ion implantation; contact formation and refractory-metal sintering; glass reflow for planarization; oxidation and titanium nitride formation; and even diffusion and gettering.

Gate dielectric formation is a difficult task because uniformity is of primary importance in these ever-thinner films. RTP is able to sequence oxidation and nitridation with sharp demarcations, while furnaces with their high thermal inertia cannot. These thin layers of various dielectric materials - oxides, nitrides, oxynitrides, etc. - can be designed to create better characteristics than oxides alone. Improvements in RTP systems have brought the uniformity of these films close to the best furnace oxidation performance.

Basic RTP heating in combination with CVD extends this layering technique and permits the sequencing of these steps in situ in a chamber or through several chambers of a cluster tool - all without exposure to external particles. A gate formation, for example, may involve the following steps:

 Preclean and deoxidize. RTP may be used as part of the preclean process.

 Oxidation and nitridation. Typical RTP steps.

 Oxide CVD. This step serves to bring the dielectric to the required thickness. It may employ RTCVD or conventional LPCVD or PECVD.

 Polysilicon CVD. RTCVD could be the best solution for this step.

 Tungsten silicide. This step serves to reduce the resistivity of polysilicon. Tungsten can be deposited by PVD, but the more practical way is by LPCVD. RTP will speed the formation of the silicide.

Figure 2. Market shares of various basic thermal systems.

Market development

The foregoing discussion provides a good background for determining and projecting the markets of these three basic thermal systems. The basic thermal equipment market from 1995 to 2000 is summarized in the table.

The basic thermal equipment market is projected to have a healthy average annual growth rate of over 15% from 1995 to 2000 in spite of its moderate performance in 1996 and 1997. By the end of the century, conventional furnaces will still have a dominant share of this market (Fig. 2). RTP will increase its share from about 16% to 23% with a strong growth rate. Fast-ramp furnaces will go from a negligible position in 1995 to nearly 9% of the total basic thermal market in 2000, with a correspondingly lush growth rate.

After 2000, the shift away from full-size furnaces is likely to accelerate, prompted partly by the proliferation of 300-mm wafers. Large batches will also have greater difficulty in meeting the finer geometries of the semiconductor roadmap.

In summary, the basic thermal equipment market is attractive. The conventional furnace makers are already protecting their growth by developing fast-ramp, minibatch versions of their equipment, which is a relatively easy technical shift. On the other hand, RTP remains a separate technology, and the RTP companies are generally not furnace makers. One can easily predict that some furnace makers will further protect their flanks by acquiring RTP technology by either license or acquisition.

John Salzer is president of Salzer Technology Enterprises Inc., 909 Berkeley St., Santa Monica, CA 90403; ph 310/828-9628, fax 310/828-9386.